4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
24 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
25 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2013 Saso Kiselkov. All rights reserved.
28 * Copyright (c) 2014 Integros [integros.com]
29 * Copyright (c) 2017 Datto Inc.
32 #include <sys/zfs_context.h>
33 #include <sys/spa_impl.h>
34 #include <sys/spa_boot.h>
36 #include <sys/zio_checksum.h>
37 #include <sys/zio_compress.h>
39 #include <sys/dmu_tx.h>
42 #include <sys/vdev_impl.h>
43 #include <sys/vdev_file.h>
44 #include <sys/metaslab.h>
45 #include <sys/uberblock_impl.h>
48 #include <sys/unique.h>
49 #include <sys/dsl_pool.h>
50 #include <sys/dsl_dir.h>
51 #include <sys/dsl_prop.h>
52 #include <sys/dsl_scan.h>
53 #include <sys/fs/zfs.h>
54 #include <sys/metaslab_impl.h>
58 #include <sys/zfeature.h>
60 #if defined(__FreeBSD__) && defined(_KERNEL)
61 #include <sys/types.h>
62 #include <sys/sysctl.h>
68 * There are four basic locks for managing spa_t structures:
70 * spa_namespace_lock (global mutex)
72 * This lock must be acquired to do any of the following:
74 * - Lookup a spa_t by name
75 * - Add or remove a spa_t from the namespace
76 * - Increase spa_refcount from non-zero
77 * - Check if spa_refcount is zero
79 * - add/remove/attach/detach devices
80 * - Held for the duration of create/destroy/import/export
82 * It does not need to handle recursion. A create or destroy may
83 * reference objects (files or zvols) in other pools, but by
84 * definition they must have an existing reference, and will never need
85 * to lookup a spa_t by name.
87 * spa_refcount (per-spa refcount_t protected by mutex)
89 * This reference count keep track of any active users of the spa_t. The
90 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
91 * the refcount is never really 'zero' - opening a pool implicitly keeps
92 * some references in the DMU. Internally we check against spa_minref, but
93 * present the image of a zero/non-zero value to consumers.
95 * spa_config_lock[] (per-spa array of rwlocks)
97 * This protects the spa_t from config changes, and must be held in
98 * the following circumstances:
100 * - RW_READER to perform I/O to the spa
101 * - RW_WRITER to change the vdev config
103 * The locking order is fairly straightforward:
105 * spa_namespace_lock -> spa_refcount
107 * The namespace lock must be acquired to increase the refcount from 0
108 * or to check if it is zero.
110 * spa_refcount -> spa_config_lock[]
112 * There must be at least one valid reference on the spa_t to acquire
115 * spa_namespace_lock -> spa_config_lock[]
117 * The namespace lock must always be taken before the config lock.
120 * The spa_namespace_lock can be acquired directly and is globally visible.
122 * The namespace is manipulated using the following functions, all of which
123 * require the spa_namespace_lock to be held.
125 * spa_lookup() Lookup a spa_t by name.
127 * spa_add() Create a new spa_t in the namespace.
129 * spa_remove() Remove a spa_t from the namespace. This also
130 * frees up any memory associated with the spa_t.
132 * spa_next() Returns the next spa_t in the system, or the
133 * first if NULL is passed.
135 * spa_evict_all() Shutdown and remove all spa_t structures in
138 * spa_guid_exists() Determine whether a pool/device guid exists.
140 * The spa_refcount is manipulated using the following functions:
142 * spa_open_ref() Adds a reference to the given spa_t. Must be
143 * called with spa_namespace_lock held if the
144 * refcount is currently zero.
146 * spa_close() Remove a reference from the spa_t. This will
147 * not free the spa_t or remove it from the
148 * namespace. No locking is required.
150 * spa_refcount_zero() Returns true if the refcount is currently
151 * zero. Must be called with spa_namespace_lock
154 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
155 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
156 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
158 * To read the configuration, it suffices to hold one of these locks as reader.
159 * To modify the configuration, you must hold all locks as writer. To modify
160 * vdev state without altering the vdev tree's topology (e.g. online/offline),
161 * you must hold SCL_STATE and SCL_ZIO as writer.
163 * We use these distinct config locks to avoid recursive lock entry.
164 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
165 * block allocations (SCL_ALLOC), which may require reading space maps
166 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
168 * The spa config locks cannot be normal rwlocks because we need the
169 * ability to hand off ownership. For example, SCL_ZIO is acquired
170 * by the issuing thread and later released by an interrupt thread.
171 * They do, however, obey the usual write-wanted semantics to prevent
172 * writer (i.e. system administrator) starvation.
174 * The lock acquisition rules are as follows:
177 * Protects changes to the vdev tree topology, such as vdev
178 * add/remove/attach/detach. Protects the dirty config list
179 * (spa_config_dirty_list) and the set of spares and l2arc devices.
182 * Protects changes to pool state and vdev state, such as vdev
183 * online/offline/fault/degrade/clear. Protects the dirty state list
184 * (spa_state_dirty_list) and global pool state (spa_state).
187 * Protects changes to metaslab groups and classes.
188 * Held as reader by metaslab_alloc() and metaslab_claim().
191 * Held by bp-level zios (those which have no io_vd upon entry)
192 * to prevent changes to the vdev tree. The bp-level zio implicitly
193 * protects all of its vdev child zios, which do not hold SCL_ZIO.
196 * Protects changes to metaslab groups and classes.
197 * Held as reader by metaslab_free(). SCL_FREE is distinct from
198 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
199 * blocks in zio_done() while another i/o that holds either
200 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
203 * Held as reader to prevent changes to the vdev tree during trivial
204 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
205 * other locks, and lower than all of them, to ensure that it's safe
206 * to acquire regardless of caller context.
208 * In addition, the following rules apply:
210 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
211 * The lock ordering is SCL_CONFIG > spa_props_lock.
213 * (b) I/O operations on leaf vdevs. For any zio operation that takes
214 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
215 * or zio_write_phys() -- the caller must ensure that the config cannot
216 * cannot change in the interim, and that the vdev cannot be reopened.
217 * SCL_STATE as reader suffices for both.
219 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
221 * spa_vdev_enter() Acquire the namespace lock and the config lock
224 * spa_vdev_exit() Release the config lock, wait for all I/O
225 * to complete, sync the updated configs to the
226 * cache, and release the namespace lock.
228 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
229 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
230 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
232 * spa_rename() is also implemented within this file since it requires
233 * manipulation of the namespace.
236 static avl_tree_t spa_namespace_avl;
237 kmutex_t spa_namespace_lock;
238 static kcondvar_t spa_namespace_cv;
239 static int spa_active_count;
240 int spa_max_replication_override = SPA_DVAS_PER_BP;
242 static kmutex_t spa_spare_lock;
243 static avl_tree_t spa_spare_avl;
244 static kmutex_t spa_l2cache_lock;
245 static avl_tree_t spa_l2cache_avl;
247 kmem_cache_t *spa_buffer_pool;
252 * Everything except dprintf, spa, and indirect_remap is on by default
255 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA | ZFS_DEBUG_INDIRECT_REMAP);
261 * zfs_recover can be set to nonzero to attempt to recover from
262 * otherwise-fatal errors, typically caused by on-disk corruption. When
263 * set, calls to zfs_panic_recover() will turn into warning messages.
264 * This should only be used as a last resort, as it typically results
265 * in leaked space, or worse.
267 boolean_t zfs_recover = B_FALSE;
270 * If destroy encounters an EIO while reading metadata (e.g. indirect
271 * blocks), space referenced by the missing metadata can not be freed.
272 * Normally this causes the background destroy to become "stalled", as
273 * it is unable to make forward progress. While in this stalled state,
274 * all remaining space to free from the error-encountering filesystem is
275 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
276 * permanently leak the space from indirect blocks that can not be read,
277 * and continue to free everything else that it can.
279 * The default, "stalling" behavior is useful if the storage partially
280 * fails (i.e. some but not all i/os fail), and then later recovers. In
281 * this case, we will be able to continue pool operations while it is
282 * partially failed, and when it recovers, we can continue to free the
283 * space, with no leaks. However, note that this case is actually
286 * Typically pools either (a) fail completely (but perhaps temporarily,
287 * e.g. a top-level vdev going offline), or (b) have localized,
288 * permanent errors (e.g. disk returns the wrong data due to bit flip or
289 * firmware bug). In case (a), this setting does not matter because the
290 * pool will be suspended and the sync thread will not be able to make
291 * forward progress regardless. In case (b), because the error is
292 * permanent, the best we can do is leak the minimum amount of space,
293 * which is what setting this flag will do. Therefore, it is reasonable
294 * for this flag to normally be set, but we chose the more conservative
295 * approach of not setting it, so that there is no possibility of
296 * leaking space in the "partial temporary" failure case.
298 boolean_t zfs_free_leak_on_eio = B_FALSE;
301 * Expiration time in milliseconds. This value has two meanings. First it is
302 * used to determine when the spa_deadman() logic should fire. By default the
303 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
304 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
305 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
308 uint64_t zfs_deadman_synctime_ms = 1000000ULL;
311 * Check time in milliseconds. This defines the frequency at which we check
314 uint64_t zfs_deadman_checktime_ms = 5000ULL;
317 * Default value of -1 for zfs_deadman_enabled is resolved in
320 int zfs_deadman_enabled = -1;
323 * The worst case is single-sector max-parity RAID-Z blocks, in which
324 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
325 * times the size; so just assume that. Add to this the fact that
326 * we can have up to 3 DVAs per bp, and one more factor of 2 because
327 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
329 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
331 int spa_asize_inflation = 24;
333 #if defined(__FreeBSD__) && defined(_KERNEL)
334 SYSCTL_DECL(_vfs_zfs);
335 SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RWTUN, &zfs_recover, 0,
336 "Try to recover from otherwise-fatal errors.");
339 sysctl_vfs_zfs_debug_flags(SYSCTL_HANDLER_ARGS)
344 err = sysctl_handle_int(oidp, &val, 0, req);
345 if (err != 0 || req->newptr == NULL)
349 * ZFS_DEBUG_MODIFY must be enabled prior to boot so all
350 * arc buffers in the system have the necessary additional
351 * checksum data. However, it is safe to disable at any
354 if (!(zfs_flags & ZFS_DEBUG_MODIFY))
355 val &= ~ZFS_DEBUG_MODIFY;
361 SYSCTL_PROC(_vfs_zfs, OID_AUTO, debugflags,
362 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RWTUN, 0, sizeof(int),
363 sysctl_vfs_zfs_debug_flags, "IU", "Debug flags for ZFS testing.");
365 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN,
366 &zfs_deadman_synctime_ms, 0,
367 "Stalled ZFS I/O expiration time in milliseconds");
368 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN,
369 &zfs_deadman_checktime_ms, 0,
370 "Period of checks for stalled ZFS I/O in milliseconds");
371 SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN,
372 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O");
373 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN,
374 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes");
383 * If we are not i386 or amd64 or in a virtual machine,
384 * disable ZFS deadman thread by default
386 if (zfs_deadman_enabled == -1) {
387 #if defined(__amd64__) || defined(__i386__)
388 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0;
390 zfs_deadman_enabled = 0;
395 #endif /* !illumos */
398 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
399 * the pool to be consumed. This ensures that we don't run the pool
400 * completely out of space, due to unaccounted changes (e.g. to the MOS).
401 * It also limits the worst-case time to allocate space. If we have
402 * less than this amount of free space, most ZPL operations (e.g. write,
403 * create) will return ENOSPC.
405 * Certain operations (e.g. file removal, most administrative actions) can
406 * use half the slop space. They will only return ENOSPC if less than half
407 * the slop space is free. Typically, once the pool has less than the slop
408 * space free, the user will use these operations to free up space in the pool.
409 * These are the operations that call dsl_pool_adjustedsize() with the netfree
410 * argument set to TRUE.
412 * Operations that are almost guaranteed to free up space in the absence of
413 * a pool checkpoint can use up to three quarters of the slop space
416 * A very restricted set of operations are always permitted, regardless of
417 * the amount of free space. These are the operations that call
418 * dsl_sync_task(ZFS_SPACE_CHECK_NONE). If these operations result in a net
419 * increase in the amount of space used, it is possible to run the pool
420 * completely out of space, causing it to be permanently read-only.
422 * Note that on very small pools, the slop space will be larger than
423 * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
424 * but we never allow it to be more than half the pool size.
426 * See also the comments in zfs_space_check_t.
428 int spa_slop_shift = 5;
429 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN,
431 "Shift value of reserved space (1/(2^spa_slop_shift)).");
432 uint64_t spa_min_slop = 128 * 1024 * 1024;
433 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, spa_min_slop, CTLFLAG_RWTUN,
435 "Minimal value of reserved space");
439 spa_load_failed(spa_t *spa, const char *fmt, ...)
445 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
448 zfs_dbgmsg("spa_load(%s, config %s): FAILED: %s", spa->spa_name,
449 spa->spa_trust_config ? "trusted" : "untrusted", buf);
454 spa_load_note(spa_t *spa, const char *fmt, ...)
460 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
463 zfs_dbgmsg("spa_load(%s, config %s): %s", spa->spa_name,
464 spa->spa_trust_config ? "trusted" : "untrusted", buf);
468 * ==========================================================================
470 * ==========================================================================
473 spa_config_lock_init(spa_t *spa)
475 for (int i = 0; i < SCL_LOCKS; i++) {
476 spa_config_lock_t *scl = &spa->spa_config_lock[i];
477 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
478 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
479 refcount_create_untracked(&scl->scl_count);
480 scl->scl_writer = NULL;
481 scl->scl_write_wanted = 0;
486 spa_config_lock_destroy(spa_t *spa)
488 for (int i = 0; i < SCL_LOCKS; i++) {
489 spa_config_lock_t *scl = &spa->spa_config_lock[i];
490 mutex_destroy(&scl->scl_lock);
491 cv_destroy(&scl->scl_cv);
492 refcount_destroy(&scl->scl_count);
493 ASSERT(scl->scl_writer == NULL);
494 ASSERT(scl->scl_write_wanted == 0);
499 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
501 for (int i = 0; i < SCL_LOCKS; i++) {
502 spa_config_lock_t *scl = &spa->spa_config_lock[i];
503 if (!(locks & (1 << i)))
505 mutex_enter(&scl->scl_lock);
506 if (rw == RW_READER) {
507 if (scl->scl_writer || scl->scl_write_wanted) {
508 mutex_exit(&scl->scl_lock);
509 spa_config_exit(spa, locks & ((1 << i) - 1),
514 ASSERT(scl->scl_writer != curthread);
515 if (!refcount_is_zero(&scl->scl_count)) {
516 mutex_exit(&scl->scl_lock);
517 spa_config_exit(spa, locks & ((1 << i) - 1),
521 scl->scl_writer = curthread;
523 (void) refcount_add(&scl->scl_count, tag);
524 mutex_exit(&scl->scl_lock);
530 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
534 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
536 for (int i = 0; i < SCL_LOCKS; i++) {
537 spa_config_lock_t *scl = &spa->spa_config_lock[i];
538 if (scl->scl_writer == curthread)
539 wlocks_held |= (1 << i);
540 if (!(locks & (1 << i)))
542 mutex_enter(&scl->scl_lock);
543 if (rw == RW_READER) {
544 while (scl->scl_writer || scl->scl_write_wanted) {
545 cv_wait(&scl->scl_cv, &scl->scl_lock);
548 ASSERT(scl->scl_writer != curthread);
549 while (!refcount_is_zero(&scl->scl_count)) {
550 scl->scl_write_wanted++;
551 cv_wait(&scl->scl_cv, &scl->scl_lock);
552 scl->scl_write_wanted--;
554 scl->scl_writer = curthread;
556 (void) refcount_add(&scl->scl_count, tag);
557 mutex_exit(&scl->scl_lock);
559 ASSERT3U(wlocks_held, <=, locks);
563 spa_config_exit(spa_t *spa, int locks, void *tag)
565 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
566 spa_config_lock_t *scl = &spa->spa_config_lock[i];
567 if (!(locks & (1 << i)))
569 mutex_enter(&scl->scl_lock);
570 ASSERT(!refcount_is_zero(&scl->scl_count));
571 if (refcount_remove(&scl->scl_count, tag) == 0) {
572 ASSERT(scl->scl_writer == NULL ||
573 scl->scl_writer == curthread);
574 scl->scl_writer = NULL; /* OK in either case */
575 cv_broadcast(&scl->scl_cv);
577 mutex_exit(&scl->scl_lock);
582 spa_config_held(spa_t *spa, int locks, krw_t rw)
586 for (int i = 0; i < SCL_LOCKS; i++) {
587 spa_config_lock_t *scl = &spa->spa_config_lock[i];
588 if (!(locks & (1 << i)))
590 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
591 (rw == RW_WRITER && scl->scl_writer == curthread))
592 locks_held |= 1 << i;
599 * ==========================================================================
600 * SPA namespace functions
601 * ==========================================================================
605 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
606 * Returns NULL if no matching spa_t is found.
609 spa_lookup(const char *name)
611 static spa_t search; /* spa_t is large; don't allocate on stack */
616 ASSERT(MUTEX_HELD(&spa_namespace_lock));
618 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
621 * If it's a full dataset name, figure out the pool name and
624 cp = strpbrk(search.spa_name, "/@#");
628 spa = avl_find(&spa_namespace_avl, &search, &where);
634 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
635 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
636 * looking for potentially hung I/Os.
639 spa_deadman(void *arg, int pending)
644 * Disable the deadman timer if the pool is suspended.
646 if (spa_suspended(spa)) {
648 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
650 /* Nothing. just don't schedule any future callouts. */
655 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
656 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
657 ++spa->spa_deadman_calls);
658 if (zfs_deadman_enabled)
659 vdev_deadman(spa->spa_root_vdev);
662 callout_schedule(&spa->spa_deadman_cycid,
663 hz * zfs_deadman_checktime_ms / MILLISEC);
668 #if defined(__FreeBSD__) && defined(_KERNEL)
670 spa_deadman_timeout(void *arg)
674 taskqueue_enqueue(taskqueue_thread, &spa->spa_deadman_task);
679 * Create an uninitialized spa_t with the given name. Requires
680 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
681 * exist by calling spa_lookup() first.
684 spa_add(const char *name, nvlist_t *config, const char *altroot)
687 spa_config_dirent_t *dp;
693 ASSERT(MUTEX_HELD(&spa_namespace_lock));
695 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
697 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
698 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
699 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
700 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
701 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
702 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
703 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
704 mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
705 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
706 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
707 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
708 mutex_init(&spa->spa_alloc_lock, NULL, MUTEX_DEFAULT, NULL);
710 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
711 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
712 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
713 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
714 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
716 for (int t = 0; t < TXG_SIZE; t++)
717 bplist_create(&spa->spa_free_bplist[t]);
719 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
720 spa->spa_state = POOL_STATE_UNINITIALIZED;
721 spa->spa_freeze_txg = UINT64_MAX;
722 spa->spa_final_txg = UINT64_MAX;
723 spa->spa_load_max_txg = UINT64_MAX;
725 spa->spa_proc_state = SPA_PROC_NONE;
726 spa->spa_trust_config = B_TRUE;
729 hdlr.cyh_func = spa_deadman;
731 hdlr.cyh_level = CY_LOW_LEVEL;
734 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
738 * This determines how often we need to check for hung I/Os after
739 * the cyclic has already fired. Since checking for hung I/Os is
740 * an expensive operation we don't want to check too frequently.
741 * Instead wait for 5 seconds before checking again.
743 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
744 when.cyt_when = CY_INFINITY;
745 mutex_enter(&cpu_lock);
746 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
747 mutex_exit(&cpu_lock);
751 * callout(9) does not provide a way to initialize a callout with
752 * a function and an argument, so we use callout_reset() to schedule
753 * the callout in the very distant future. Even if that event ever
754 * fires, it should be okayas we won't have any active zio-s.
755 * But normally spa_sync() will reschedule the callout with a proper
757 * callout(9) does not allow the callback function to sleep but
758 * vdev_deadman() needs to acquire vq_lock and illumos mutexes are
759 * emulated using sx(9). For this reason spa_deadman_timeout()
760 * will schedule spa_deadman() as task on a taskqueue that allows
763 TASK_INIT(&spa->spa_deadman_task, 0, spa_deadman, spa);
764 callout_init(&spa->spa_deadman_cycid, 1);
765 callout_reset_sbt(&spa->spa_deadman_cycid, SBT_MAX, 0,
766 spa_deadman_timeout, spa, 0);
769 refcount_create(&spa->spa_refcount);
770 spa_config_lock_init(spa);
772 avl_add(&spa_namespace_avl, spa);
775 * Set the alternate root, if there is one.
778 spa->spa_root = spa_strdup(altroot);
782 avl_create(&spa->spa_alloc_tree, zio_bookmark_compare,
783 sizeof (zio_t), offsetof(zio_t, io_alloc_node));
786 * Every pool starts with the default cachefile
788 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
789 offsetof(spa_config_dirent_t, scd_link));
791 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
792 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
793 list_insert_head(&spa->spa_config_list, dp);
795 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
798 if (config != NULL) {
801 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
803 VERIFY(nvlist_dup(features, &spa->spa_label_features,
807 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
810 if (spa->spa_label_features == NULL) {
811 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
815 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
817 spa->spa_min_ashift = INT_MAX;
818 spa->spa_max_ashift = 0;
821 * As a pool is being created, treat all features as disabled by
822 * setting SPA_FEATURE_DISABLED for all entries in the feature
825 for (int i = 0; i < SPA_FEATURES; i++) {
826 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
833 * Removes a spa_t from the namespace, freeing up any memory used. Requires
834 * spa_namespace_lock. This is called only after the spa_t has been closed and
838 spa_remove(spa_t *spa)
840 spa_config_dirent_t *dp;
842 ASSERT(MUTEX_HELD(&spa_namespace_lock));
843 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
844 ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
846 nvlist_free(spa->spa_config_splitting);
848 avl_remove(&spa_namespace_avl, spa);
849 cv_broadcast(&spa_namespace_cv);
852 spa_strfree(spa->spa_root);
856 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
857 list_remove(&spa->spa_config_list, dp);
858 if (dp->scd_path != NULL)
859 spa_strfree(dp->scd_path);
860 kmem_free(dp, sizeof (spa_config_dirent_t));
863 avl_destroy(&spa->spa_alloc_tree);
864 list_destroy(&spa->spa_config_list);
866 nvlist_free(spa->spa_label_features);
867 nvlist_free(spa->spa_load_info);
868 spa_config_set(spa, NULL);
871 mutex_enter(&cpu_lock);
872 if (spa->spa_deadman_cycid != CYCLIC_NONE)
873 cyclic_remove(spa->spa_deadman_cycid);
874 mutex_exit(&cpu_lock);
875 spa->spa_deadman_cycid = CYCLIC_NONE;
878 callout_drain(&spa->spa_deadman_cycid);
879 taskqueue_drain(taskqueue_thread, &spa->spa_deadman_task);
883 refcount_destroy(&spa->spa_refcount);
885 spa_config_lock_destroy(spa);
887 for (int t = 0; t < TXG_SIZE; t++)
888 bplist_destroy(&spa->spa_free_bplist[t]);
890 zio_checksum_templates_free(spa);
892 cv_destroy(&spa->spa_async_cv);
893 cv_destroy(&spa->spa_evicting_os_cv);
894 cv_destroy(&spa->spa_proc_cv);
895 cv_destroy(&spa->spa_scrub_io_cv);
896 cv_destroy(&spa->spa_suspend_cv);
898 mutex_destroy(&spa->spa_alloc_lock);
899 mutex_destroy(&spa->spa_async_lock);
900 mutex_destroy(&spa->spa_errlist_lock);
901 mutex_destroy(&spa->spa_errlog_lock);
902 mutex_destroy(&spa->spa_evicting_os_lock);
903 mutex_destroy(&spa->spa_history_lock);
904 mutex_destroy(&spa->spa_proc_lock);
905 mutex_destroy(&spa->spa_props_lock);
906 mutex_destroy(&spa->spa_cksum_tmpls_lock);
907 mutex_destroy(&spa->spa_scrub_lock);
908 mutex_destroy(&spa->spa_suspend_lock);
909 mutex_destroy(&spa->spa_vdev_top_lock);
911 kmem_free(spa, sizeof (spa_t));
915 * Given a pool, return the next pool in the namespace, or NULL if there is
916 * none. If 'prev' is NULL, return the first pool.
919 spa_next(spa_t *prev)
921 ASSERT(MUTEX_HELD(&spa_namespace_lock));
924 return (AVL_NEXT(&spa_namespace_avl, prev));
926 return (avl_first(&spa_namespace_avl));
930 * ==========================================================================
931 * SPA refcount functions
932 * ==========================================================================
936 * Add a reference to the given spa_t. Must have at least one reference, or
937 * have the namespace lock held.
940 spa_open_ref(spa_t *spa, void *tag)
942 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
943 MUTEX_HELD(&spa_namespace_lock));
944 (void) refcount_add(&spa->spa_refcount, tag);
948 * Remove a reference to the given spa_t. Must have at least one reference, or
949 * have the namespace lock held.
952 spa_close(spa_t *spa, void *tag)
954 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
955 MUTEX_HELD(&spa_namespace_lock));
956 (void) refcount_remove(&spa->spa_refcount, tag);
960 * Remove a reference to the given spa_t held by a dsl dir that is
961 * being asynchronously released. Async releases occur from a taskq
962 * performing eviction of dsl datasets and dirs. The namespace lock
963 * isn't held and the hold by the object being evicted may contribute to
964 * spa_minref (e.g. dataset or directory released during pool export),
965 * so the asserts in spa_close() do not apply.
968 spa_async_close(spa_t *spa, void *tag)
970 (void) refcount_remove(&spa->spa_refcount, tag);
974 * Check to see if the spa refcount is zero. Must be called with
975 * spa_namespace_lock held. We really compare against spa_minref, which is the
976 * number of references acquired when opening a pool
979 spa_refcount_zero(spa_t *spa)
981 ASSERT(MUTEX_HELD(&spa_namespace_lock));
983 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
987 * ==========================================================================
988 * SPA spare and l2cache tracking
989 * ==========================================================================
993 * Hot spares and cache devices are tracked using the same code below,
994 * for 'auxiliary' devices.
997 typedef struct spa_aux {
1005 spa_aux_compare(const void *a, const void *b)
1007 const spa_aux_t *sa = a;
1008 const spa_aux_t *sb = b;
1010 if (sa->aux_guid < sb->aux_guid)
1012 else if (sa->aux_guid > sb->aux_guid)
1019 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
1025 search.aux_guid = vd->vdev_guid;
1026 if ((aux = avl_find(avl, &search, &where)) != NULL) {
1029 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
1030 aux->aux_guid = vd->vdev_guid;
1032 avl_insert(avl, aux, where);
1037 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
1043 search.aux_guid = vd->vdev_guid;
1044 aux = avl_find(avl, &search, &where);
1046 ASSERT(aux != NULL);
1048 if (--aux->aux_count == 0) {
1049 avl_remove(avl, aux);
1050 kmem_free(aux, sizeof (spa_aux_t));
1051 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
1052 aux->aux_pool = 0ULL;
1057 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
1059 spa_aux_t search, *found;
1061 search.aux_guid = guid;
1062 found = avl_find(avl, &search, NULL);
1066 *pool = found->aux_pool;
1073 *refcnt = found->aux_count;
1078 return (found != NULL);
1082 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
1084 spa_aux_t search, *found;
1087 search.aux_guid = vd->vdev_guid;
1088 found = avl_find(avl, &search, &where);
1089 ASSERT(found != NULL);
1090 ASSERT(found->aux_pool == 0ULL);
1092 found->aux_pool = spa_guid(vd->vdev_spa);
1096 * Spares are tracked globally due to the following constraints:
1098 * - A spare may be part of multiple pools.
1099 * - A spare may be added to a pool even if it's actively in use within
1101 * - A spare in use in any pool can only be the source of a replacement if
1102 * the target is a spare in the same pool.
1104 * We keep track of all spares on the system through the use of a reference
1105 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
1106 * spare, then we bump the reference count in the AVL tree. In addition, we set
1107 * the 'vdev_isspare' member to indicate that the device is a spare (active or
1108 * inactive). When a spare is made active (used to replace a device in the
1109 * pool), we also keep track of which pool its been made a part of.
1111 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
1112 * called under the spa_namespace lock as part of vdev reconfiguration. The
1113 * separate spare lock exists for the status query path, which does not need to
1114 * be completely consistent with respect to other vdev configuration changes.
1118 spa_spare_compare(const void *a, const void *b)
1120 return (spa_aux_compare(a, b));
1124 spa_spare_add(vdev_t *vd)
1126 mutex_enter(&spa_spare_lock);
1127 ASSERT(!vd->vdev_isspare);
1128 spa_aux_add(vd, &spa_spare_avl);
1129 vd->vdev_isspare = B_TRUE;
1130 mutex_exit(&spa_spare_lock);
1134 spa_spare_remove(vdev_t *vd)
1136 mutex_enter(&spa_spare_lock);
1137 ASSERT(vd->vdev_isspare);
1138 spa_aux_remove(vd, &spa_spare_avl);
1139 vd->vdev_isspare = B_FALSE;
1140 mutex_exit(&spa_spare_lock);
1144 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1148 mutex_enter(&spa_spare_lock);
1149 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1150 mutex_exit(&spa_spare_lock);
1156 spa_spare_activate(vdev_t *vd)
1158 mutex_enter(&spa_spare_lock);
1159 ASSERT(vd->vdev_isspare);
1160 spa_aux_activate(vd, &spa_spare_avl);
1161 mutex_exit(&spa_spare_lock);
1165 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1166 * Cache devices currently only support one pool per cache device, and so
1167 * for these devices the aux reference count is currently unused beyond 1.
1171 spa_l2cache_compare(const void *a, const void *b)
1173 return (spa_aux_compare(a, b));
1177 spa_l2cache_add(vdev_t *vd)
1179 mutex_enter(&spa_l2cache_lock);
1180 ASSERT(!vd->vdev_isl2cache);
1181 spa_aux_add(vd, &spa_l2cache_avl);
1182 vd->vdev_isl2cache = B_TRUE;
1183 mutex_exit(&spa_l2cache_lock);
1187 spa_l2cache_remove(vdev_t *vd)
1189 mutex_enter(&spa_l2cache_lock);
1190 ASSERT(vd->vdev_isl2cache);
1191 spa_aux_remove(vd, &spa_l2cache_avl);
1192 vd->vdev_isl2cache = B_FALSE;
1193 mutex_exit(&spa_l2cache_lock);
1197 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1201 mutex_enter(&spa_l2cache_lock);
1202 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1203 mutex_exit(&spa_l2cache_lock);
1209 spa_l2cache_activate(vdev_t *vd)
1211 mutex_enter(&spa_l2cache_lock);
1212 ASSERT(vd->vdev_isl2cache);
1213 spa_aux_activate(vd, &spa_l2cache_avl);
1214 mutex_exit(&spa_l2cache_lock);
1218 * ==========================================================================
1220 * ==========================================================================
1224 * Lock the given spa_t for the purpose of adding or removing a vdev.
1225 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1226 * It returns the next transaction group for the spa_t.
1229 spa_vdev_enter(spa_t *spa)
1231 mutex_enter(&spa->spa_vdev_top_lock);
1232 mutex_enter(&spa_namespace_lock);
1233 return (spa_vdev_config_enter(spa));
1237 * Internal implementation for spa_vdev_enter(). Used when a vdev
1238 * operation requires multiple syncs (i.e. removing a device) while
1239 * keeping the spa_namespace_lock held.
1242 spa_vdev_config_enter(spa_t *spa)
1244 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1246 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1248 return (spa_last_synced_txg(spa) + 1);
1252 * Used in combination with spa_vdev_config_enter() to allow the syncing
1253 * of multiple transactions without releasing the spa_namespace_lock.
1256 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1258 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1260 int config_changed = B_FALSE;
1262 ASSERT(txg > spa_last_synced_txg(spa));
1264 spa->spa_pending_vdev = NULL;
1267 * Reassess the DTLs.
1269 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1271 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1272 config_changed = B_TRUE;
1273 spa->spa_config_generation++;
1277 * Verify the metaslab classes.
1279 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1280 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1282 spa_config_exit(spa, SCL_ALL, spa);
1285 * Panic the system if the specified tag requires it. This
1286 * is useful for ensuring that configurations are updated
1289 if (zio_injection_enabled)
1290 zio_handle_panic_injection(spa, tag, 0);
1293 * Note: this txg_wait_synced() is important because it ensures
1294 * that there won't be more than one config change per txg.
1295 * This allows us to use the txg as the generation number.
1298 txg_wait_synced(spa->spa_dsl_pool, txg);
1301 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1302 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1304 spa_config_exit(spa, SCL_ALL, spa);
1308 * If the config changed, update the config cache.
1311 spa_write_cachefile(spa, B_FALSE, B_TRUE);
1315 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1316 * locking of spa_vdev_enter(), we also want make sure the transactions have
1317 * synced to disk, and then update the global configuration cache with the new
1321 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1323 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1324 mutex_exit(&spa_namespace_lock);
1325 mutex_exit(&spa->spa_vdev_top_lock);
1331 * Lock the given spa_t for the purpose of changing vdev state.
1334 spa_vdev_state_enter(spa_t *spa, int oplocks)
1336 int locks = SCL_STATE_ALL | oplocks;
1339 * Root pools may need to read of the underlying devfs filesystem
1340 * when opening up a vdev. Unfortunately if we're holding the
1341 * SCL_ZIO lock it will result in a deadlock when we try to issue
1342 * the read from the root filesystem. Instead we "prefetch"
1343 * the associated vnodes that we need prior to opening the
1344 * underlying devices and cache them so that we can prevent
1345 * any I/O when we are doing the actual open.
1347 if (spa_is_root(spa)) {
1348 int low = locks & ~(SCL_ZIO - 1);
1349 int high = locks & ~low;
1351 spa_config_enter(spa, high, spa, RW_WRITER);
1352 vdev_hold(spa->spa_root_vdev);
1353 spa_config_enter(spa, low, spa, RW_WRITER);
1355 spa_config_enter(spa, locks, spa, RW_WRITER);
1357 spa->spa_vdev_locks = locks;
1361 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1363 boolean_t config_changed = B_FALSE;
1365 if (vd != NULL || error == 0)
1366 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1370 vdev_state_dirty(vd->vdev_top);
1371 config_changed = B_TRUE;
1372 spa->spa_config_generation++;
1375 if (spa_is_root(spa))
1376 vdev_rele(spa->spa_root_vdev);
1378 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1379 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1382 * If anything changed, wait for it to sync. This ensures that,
1383 * from the system administrator's perspective, zpool(1M) commands
1384 * are synchronous. This is important for things like zpool offline:
1385 * when the command completes, you expect no further I/O from ZFS.
1388 txg_wait_synced(spa->spa_dsl_pool, 0);
1391 * If the config changed, update the config cache.
1393 if (config_changed) {
1394 mutex_enter(&spa_namespace_lock);
1395 spa_write_cachefile(spa, B_FALSE, B_TRUE);
1396 mutex_exit(&spa_namespace_lock);
1403 * ==========================================================================
1404 * Miscellaneous functions
1405 * ==========================================================================
1409 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1411 if (!nvlist_exists(spa->spa_label_features, feature)) {
1412 fnvlist_add_boolean(spa->spa_label_features, feature);
1414 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1415 * dirty the vdev config because lock SCL_CONFIG is not held.
1416 * Thankfully, in this case we don't need to dirty the config
1417 * because it will be written out anyway when we finish
1418 * creating the pool.
1420 if (tx->tx_txg != TXG_INITIAL)
1421 vdev_config_dirty(spa->spa_root_vdev);
1426 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1428 if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1429 vdev_config_dirty(spa->spa_root_vdev);
1436 spa_rename(const char *name, const char *newname)
1442 * Lookup the spa_t and grab the config lock for writing. We need to
1443 * actually open the pool so that we can sync out the necessary labels.
1444 * It's OK to call spa_open() with the namespace lock held because we
1445 * allow recursive calls for other reasons.
1447 mutex_enter(&spa_namespace_lock);
1448 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1449 mutex_exit(&spa_namespace_lock);
1453 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1455 avl_remove(&spa_namespace_avl, spa);
1456 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1457 avl_add(&spa_namespace_avl, spa);
1460 * Sync all labels to disk with the new names by marking the root vdev
1461 * dirty and waiting for it to sync. It will pick up the new pool name
1464 vdev_config_dirty(spa->spa_root_vdev);
1466 spa_config_exit(spa, SCL_ALL, FTAG);
1468 txg_wait_synced(spa->spa_dsl_pool, 0);
1471 * Sync the updated config cache.
1473 spa_write_cachefile(spa, B_FALSE, B_TRUE);
1475 spa_close(spa, FTAG);
1477 mutex_exit(&spa_namespace_lock);
1483 * Return the spa_t associated with given pool_guid, if it exists. If
1484 * device_guid is non-zero, determine whether the pool exists *and* contains
1485 * a device with the specified device_guid.
1488 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1491 avl_tree_t *t = &spa_namespace_avl;
1493 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1495 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1496 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1498 if (spa->spa_root_vdev == NULL)
1500 if (spa_guid(spa) == pool_guid) {
1501 if (device_guid == 0)
1504 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1505 device_guid) != NULL)
1509 * Check any devices we may be in the process of adding.
1511 if (spa->spa_pending_vdev) {
1512 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1513 device_guid) != NULL)
1523 * Determine whether a pool with the given pool_guid exists.
1526 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1528 return (spa_by_guid(pool_guid, device_guid) != NULL);
1532 spa_strdup(const char *s)
1538 new = kmem_alloc(len + 1, KM_SLEEP);
1546 spa_strfree(char *s)
1548 kmem_free(s, strlen(s) + 1);
1552 spa_get_random(uint64_t range)
1558 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1564 spa_generate_guid(spa_t *spa)
1566 uint64_t guid = spa_get_random(-1ULL);
1569 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1570 guid = spa_get_random(-1ULL);
1572 while (guid == 0 || spa_guid_exists(guid, 0))
1573 guid = spa_get_random(-1ULL);
1580 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1583 char *checksum = NULL;
1584 char *compress = NULL;
1587 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1588 dmu_object_byteswap_t bswap =
1589 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1590 (void) snprintf(type, sizeof (type), "bswap %s %s",
1591 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1592 "metadata" : "data",
1593 dmu_ot_byteswap[bswap].ob_name);
1595 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1598 if (!BP_IS_EMBEDDED(bp)) {
1600 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1602 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1605 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1610 spa_freeze(spa_t *spa)
1612 uint64_t freeze_txg = 0;
1614 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1615 if (spa->spa_freeze_txg == UINT64_MAX) {
1616 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1617 spa->spa_freeze_txg = freeze_txg;
1619 spa_config_exit(spa, SCL_ALL, FTAG);
1620 if (freeze_txg != 0)
1621 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1625 zfs_panic_recover(const char *fmt, ...)
1630 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1635 * This is a stripped-down version of strtoull, suitable only for converting
1636 * lowercase hexadecimal numbers that don't overflow.
1639 zfs_strtonum(const char *str, char **nptr)
1645 while ((c = *str) != '\0') {
1646 if (c >= '0' && c <= '9')
1648 else if (c >= 'a' && c <= 'f')
1649 digit = 10 + c - 'a';
1660 *nptr = (char *)str;
1666 * ==========================================================================
1667 * Accessor functions
1668 * ==========================================================================
1672 spa_shutting_down(spa_t *spa)
1674 return (spa->spa_async_suspended);
1678 spa_get_dsl(spa_t *spa)
1680 return (spa->spa_dsl_pool);
1684 spa_is_initializing(spa_t *spa)
1686 return (spa->spa_is_initializing);
1690 spa_indirect_vdevs_loaded(spa_t *spa)
1692 return (spa->spa_indirect_vdevs_loaded);
1696 spa_get_rootblkptr(spa_t *spa)
1698 return (&spa->spa_ubsync.ub_rootbp);
1702 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1704 spa->spa_uberblock.ub_rootbp = *bp;
1708 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1710 if (spa->spa_root == NULL)
1713 (void) strncpy(buf, spa->spa_root, buflen);
1717 spa_sync_pass(spa_t *spa)
1719 return (spa->spa_sync_pass);
1723 spa_name(spa_t *spa)
1725 return (spa->spa_name);
1729 spa_guid(spa_t *spa)
1731 dsl_pool_t *dp = spa_get_dsl(spa);
1735 * If we fail to parse the config during spa_load(), we can go through
1736 * the error path (which posts an ereport) and end up here with no root
1737 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1740 if (spa->spa_root_vdev == NULL)
1741 return (spa->spa_config_guid);
1743 guid = spa->spa_last_synced_guid != 0 ?
1744 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1747 * Return the most recently synced out guid unless we're
1748 * in syncing context.
1750 if (dp && dsl_pool_sync_context(dp))
1751 return (spa->spa_root_vdev->vdev_guid);
1757 spa_load_guid(spa_t *spa)
1760 * This is a GUID that exists solely as a reference for the
1761 * purposes of the arc. It is generated at load time, and
1762 * is never written to persistent storage.
1764 return (spa->spa_load_guid);
1768 spa_last_synced_txg(spa_t *spa)
1770 return (spa->spa_ubsync.ub_txg);
1774 spa_first_txg(spa_t *spa)
1776 return (spa->spa_first_txg);
1780 spa_syncing_txg(spa_t *spa)
1782 return (spa->spa_syncing_txg);
1786 * Return the last txg where data can be dirtied. The final txgs
1787 * will be used to just clear out any deferred frees that remain.
1790 spa_final_dirty_txg(spa_t *spa)
1792 return (spa->spa_final_txg - TXG_DEFER_SIZE);
1796 spa_state(spa_t *spa)
1798 return (spa->spa_state);
1802 spa_load_state(spa_t *spa)
1804 return (spa->spa_load_state);
1808 spa_freeze_txg(spa_t *spa)
1810 return (spa->spa_freeze_txg);
1815 spa_get_worst_case_asize(spa_t *spa, uint64_t lsize)
1817 return (lsize * spa_asize_inflation);
1821 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1822 * or at least 128MB, unless that would cause it to be more than half the
1825 * See the comment above spa_slop_shift for details.
1828 spa_get_slop_space(spa_t *spa)
1830 uint64_t space = spa_get_dspace(spa);
1831 return (MAX(space >> spa_slop_shift, MIN(space >> 1, spa_min_slop)));
1835 spa_get_dspace(spa_t *spa)
1837 return (spa->spa_dspace);
1841 spa_get_checkpoint_space(spa_t *spa)
1843 return (spa->spa_checkpoint_info.sci_dspace);
1847 spa_update_dspace(spa_t *spa)
1849 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1850 ddt_get_dedup_dspace(spa);
1851 if (spa->spa_vdev_removal != NULL) {
1853 * We can't allocate from the removing device, so
1854 * subtract its size. This prevents the DMU/DSL from
1855 * filling up the (now smaller) pool while we are in the
1856 * middle of removing the device.
1858 * Note that the DMU/DSL doesn't actually know or care
1859 * how much space is allocated (it does its own tracking
1860 * of how much space has been logically used). So it
1861 * doesn't matter that the data we are moving may be
1862 * allocated twice (on the old device and the new
1865 vdev_t *vd = spa->spa_vdev_removal->svr_vdev;
1866 spa->spa_dspace -= spa_deflate(spa) ?
1867 vd->vdev_stat.vs_dspace : vd->vdev_stat.vs_space;
1872 * Return the failure mode that has been set to this pool. The default
1873 * behavior will be to block all I/Os when a complete failure occurs.
1876 spa_get_failmode(spa_t *spa)
1878 return (spa->spa_failmode);
1882 spa_suspended(spa_t *spa)
1884 return (spa->spa_suspended);
1888 spa_version(spa_t *spa)
1890 return (spa->spa_ubsync.ub_version);
1894 spa_deflate(spa_t *spa)
1896 return (spa->spa_deflate);
1900 spa_normal_class(spa_t *spa)
1902 return (spa->spa_normal_class);
1906 spa_log_class(spa_t *spa)
1908 return (spa->spa_log_class);
1912 spa_evicting_os_register(spa_t *spa, objset_t *os)
1914 mutex_enter(&spa->spa_evicting_os_lock);
1915 list_insert_head(&spa->spa_evicting_os_list, os);
1916 mutex_exit(&spa->spa_evicting_os_lock);
1920 spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1922 mutex_enter(&spa->spa_evicting_os_lock);
1923 list_remove(&spa->spa_evicting_os_list, os);
1924 cv_broadcast(&spa->spa_evicting_os_cv);
1925 mutex_exit(&spa->spa_evicting_os_lock);
1929 spa_evicting_os_wait(spa_t *spa)
1931 mutex_enter(&spa->spa_evicting_os_lock);
1932 while (!list_is_empty(&spa->spa_evicting_os_list))
1933 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1934 mutex_exit(&spa->spa_evicting_os_lock);
1936 dmu_buf_user_evict_wait();
1940 spa_max_replication(spa_t *spa)
1943 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1944 * handle BPs with more than one DVA allocated. Set our max
1945 * replication level accordingly.
1947 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1949 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1953 spa_prev_software_version(spa_t *spa)
1955 return (spa->spa_prev_software_version);
1959 spa_deadman_synctime(spa_t *spa)
1961 return (spa->spa_deadman_synctime);
1965 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1967 uint64_t asize = DVA_GET_ASIZE(dva);
1968 uint64_t dsize = asize;
1970 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1972 if (asize != 0 && spa->spa_deflate) {
1973 uint64_t vdev = DVA_GET_VDEV(dva);
1974 vdev_t *vd = vdev_lookup_top(spa, vdev);
1977 "dva_get_dsize_sync(): bad DVA %llu:%llu",
1978 (u_longlong_t)vdev, (u_longlong_t)asize);
1980 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1987 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1991 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1992 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1998 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
2002 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2004 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
2005 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
2007 spa_config_exit(spa, SCL_VDEV, FTAG);
2013 * ==========================================================================
2014 * Initialization and Termination
2015 * ==========================================================================
2019 spa_name_compare(const void *a1, const void *a2)
2021 const spa_t *s1 = a1;
2022 const spa_t *s2 = a2;
2025 s = strcmp(s1->spa_name, s2->spa_name);
2036 return (spa_active_count);
2046 EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0);
2052 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
2053 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
2054 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
2055 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
2057 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
2058 offsetof(spa_t, spa_avl));
2060 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
2061 offsetof(spa_aux_t, aux_avl));
2063 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
2064 offsetof(spa_aux_t, aux_avl));
2066 spa_mode_global = mode;
2072 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
2073 arc_procfd = open("/proc/self/ctl", O_WRONLY);
2074 if (arc_procfd == -1) {
2075 perror("could not enable watchpoints: "
2076 "opening /proc/self/ctl failed: ");
2082 #endif /* illumos */
2086 metaslab_alloc_trace_init();
2091 vdev_cache_stat_init();
2095 zpool_feature_init();
2102 #endif /* !illumos */
2113 vdev_cache_stat_fini();
2118 metaslab_alloc_trace_fini();
2123 avl_destroy(&spa_namespace_avl);
2124 avl_destroy(&spa_spare_avl);
2125 avl_destroy(&spa_l2cache_avl);
2127 cv_destroy(&spa_namespace_cv);
2128 mutex_destroy(&spa_namespace_lock);
2129 mutex_destroy(&spa_spare_lock);
2130 mutex_destroy(&spa_l2cache_lock);
2134 * Return whether this pool has slogs. No locking needed.
2135 * It's not a problem if the wrong answer is returned as it's only for
2136 * performance and not correctness
2139 spa_has_slogs(spa_t *spa)
2141 return (spa->spa_log_class->mc_rotor != NULL);
2145 spa_get_log_state(spa_t *spa)
2147 return (spa->spa_log_state);
2151 spa_set_log_state(spa_t *spa, spa_log_state_t state)
2153 spa->spa_log_state = state;
2157 spa_is_root(spa_t *spa)
2159 return (spa->spa_is_root);
2163 spa_writeable(spa_t *spa)
2165 return (!!(spa->spa_mode & FWRITE) && spa->spa_trust_config);
2169 * Returns true if there is a pending sync task in any of the current
2170 * syncing txg, the current quiescing txg, or the current open txg.
2173 spa_has_pending_synctask(spa_t *spa)
2175 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks) ||
2176 !txg_all_lists_empty(&spa->spa_dsl_pool->dp_early_sync_tasks));
2180 spa_mode(spa_t *spa)
2182 return (spa->spa_mode);
2186 spa_bootfs(spa_t *spa)
2188 return (spa->spa_bootfs);
2192 spa_delegation(spa_t *spa)
2194 return (spa->spa_delegation);
2198 spa_meta_objset(spa_t *spa)
2200 return (spa->spa_meta_objset);
2204 spa_dedup_checksum(spa_t *spa)
2206 return (spa->spa_dedup_checksum);
2210 * Reset pool scan stat per scan pass (or reboot).
2213 spa_scan_stat_init(spa_t *spa)
2215 /* data not stored on disk */
2216 spa->spa_scan_pass_start = gethrestime_sec();
2217 if (dsl_scan_is_paused_scrub(spa->spa_dsl_pool->dp_scan))
2218 spa->spa_scan_pass_scrub_pause = spa->spa_scan_pass_start;
2220 spa->spa_scan_pass_scrub_pause = 0;
2221 spa->spa_scan_pass_scrub_spent_paused = 0;
2222 spa->spa_scan_pass_exam = 0;
2223 vdev_scan_stat_init(spa->spa_root_vdev);
2227 * Get scan stats for zpool status reports
2230 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2232 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2234 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2235 return (SET_ERROR(ENOENT));
2236 bzero(ps, sizeof (pool_scan_stat_t));
2238 /* data stored on disk */
2239 ps->pss_func = scn->scn_phys.scn_func;
2240 ps->pss_start_time = scn->scn_phys.scn_start_time;
2241 ps->pss_end_time = scn->scn_phys.scn_end_time;
2242 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2243 ps->pss_examined = scn->scn_phys.scn_examined;
2244 ps->pss_to_process = scn->scn_phys.scn_to_process;
2245 ps->pss_processed = scn->scn_phys.scn_processed;
2246 ps->pss_errors = scn->scn_phys.scn_errors;
2247 ps->pss_state = scn->scn_phys.scn_state;
2249 /* data not stored on disk */
2250 ps->pss_pass_start = spa->spa_scan_pass_start;
2251 ps->pss_pass_exam = spa->spa_scan_pass_exam;
2252 ps->pss_pass_scrub_pause = spa->spa_scan_pass_scrub_pause;
2253 ps->pss_pass_scrub_spent_paused = spa->spa_scan_pass_scrub_spent_paused;
2259 spa_debug_enabled(spa_t *spa)
2261 return (spa->spa_debug);
2265 spa_maxblocksize(spa_t *spa)
2267 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2268 return (SPA_MAXBLOCKSIZE);
2270 return (SPA_OLD_MAXBLOCKSIZE);
2274 * Returns the txg that the last device removal completed. No indirect mappings
2275 * have been added since this txg.
2278 spa_get_last_removal_txg(spa_t *spa)
2281 uint64_t ret = -1ULL;
2283 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2285 * sr_prev_indirect_vdev is only modified while holding all the
2286 * config locks, so it is sufficient to hold SCL_VDEV as reader when
2289 vdevid = spa->spa_removing_phys.sr_prev_indirect_vdev;
2291 while (vdevid != -1ULL) {
2292 vdev_t *vd = vdev_lookup_top(spa, vdevid);
2293 vdev_indirect_births_t *vib = vd->vdev_indirect_births;
2295 ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
2298 * If the removal did not remap any data, we don't care.
2300 if (vdev_indirect_births_count(vib) != 0) {
2301 ret = vdev_indirect_births_last_entry_txg(vib);
2305 vdevid = vd->vdev_indirect_config.vic_prev_indirect_vdev;
2307 spa_config_exit(spa, SCL_VDEV, FTAG);
2310 spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
2316 spa_trust_config(spa_t *spa)
2318 return (spa->spa_trust_config);
2322 spa_missing_tvds_allowed(spa_t *spa)
2324 return (spa->spa_missing_tvds_allowed);
2328 spa_set_missing_tvds(spa_t *spa, uint64_t missing)
2330 spa->spa_missing_tvds = missing;
2334 spa_top_vdevs_spacemap_addressable(spa_t *spa)
2336 vdev_t *rvd = spa->spa_root_vdev;
2337 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2338 if (!vdev_is_spacemap_addressable(rvd->vdev_child[c]))
2345 spa_has_checkpoint(spa_t *spa)
2347 return (spa->spa_checkpoint_txg != 0);
2351 spa_importing_readonly_checkpoint(spa_t *spa)
2353 return ((spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT) &&
2354 spa->spa_mode == FREAD);
2358 spa_min_claim_txg(spa_t *spa)
2360 uint64_t checkpoint_txg = spa->spa_uberblock.ub_checkpoint_txg;
2362 if (checkpoint_txg != 0)
2363 return (checkpoint_txg + 1);
2365 return (spa->spa_first_txg);
2369 * If there is a checkpoint, async destroys may consume more space from
2370 * the pool instead of freeing it. In an attempt to save the pool from
2371 * getting suspended when it is about to run out of space, we stop
2372 * processing async destroys.
2375 spa_suspend_async_destroy(spa_t *spa)
2377 dsl_pool_t *dp = spa_get_dsl(spa);
2379 uint64_t unreserved = dsl_pool_unreserved_space(dp,
2380 ZFS_SPACE_CHECK_EXTRA_RESERVED);
2381 uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes;
2382 uint64_t avail = (unreserved > used) ? (unreserved - used) : 0;
2384 if (spa_has_checkpoint(spa) && avail == 0)